基于多特征和证据理论离线签名识别

摘 要 本文提出一种基于多特征和证据理论的离线签名识别方法。首先通过比较各个特征在离线签名识别中的性能，选取比较有效的平均伪动态特征和中等分辨率网格特征作为识别特征。然后构造了两个k近邻（KNN,k Nearest Neighbor）分类器，对签名图象进行初步识别。在初步识别基础上，利用一种改进的证据理论合成公式，将两个KNN分类器的输出结果进行融合得到最终识别结果。实验结果表明新的识别方法是有效性。

Off-line signature recognition based on multi-features and evidence theory

A off-line signature recognition method based on multi-features and evidence theory is introduced in this paper.Two effective features,pseudo dynamic feature and grid feature,are selected by comparing the performance of features in off-line signature recognition.Then initial signature recognition result is done by two KNN classifiers.Next,the output from classifiers is used as evidence to get the final fusing result,which using a improved combined method based on evidence theory.Experiments show that this new recognition method is effective.